Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Remove omp_get_max_threads in data. (#7588)

Browse Source
This commit is contained in:
Jiaming Yuan
2022-01-24 02:44:07 +08:00
committed by GitHub
parent f84291c1e1
commit 5817840858
18 changed files with 97 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
tests/cpp/common/test_column_matrix.cc
View File

@@ -1,3 +1,6 @@
/*!
* Copyright 2018-2022 by XGBoost Contributors
*/
#include <dmlc/filesystem.h>
#include <gtest/gtest.h>
@@ -14,7 +17,7 @@ TEST(DenseColumn, Test) {
static_cast<uint64_t>(std::numeric_limits<uint16_t>::max()) + 2};
for (size_t max_num_bin : max_num_bins) {
auto dmat = RandomDataGenerator(100, 10, 0.0).GenerateDMatrix();
GHistIndexMatrix gmat(dmat.get(), max_num_bin, false);
GHistIndexMatrix gmat(dmat.get(), max_num_bin, false, common::OmpGetNumThreads(0));
ColumnMatrix column_matrix;
column_matrix.Init(gmat, 0.2);
@@ -61,7 +64,7 @@ TEST(SparseColumn, Test) {
static_cast<uint64_t>(std::numeric_limits<uint16_t>::max()) + 2};
for (size_t max_num_bin : max_num_bins) {
auto dmat = RandomDataGenerator(100, 1, 0.85).GenerateDMatrix();
GHistIndexMatrix gmat(dmat.get(), max_num_bin, false);
GHistIndexMatrix gmat(dmat.get(), max_num_bin, false, common::OmpGetNumThreads(0));
ColumnMatrix column_matrix;
column_matrix.Init(gmat, 0.5);
switch (column_matrix.GetTypeSize()) {
@@ -101,7 +104,7 @@ TEST(DenseColumnWithMissing, Test) {
static_cast<uint64_t>(std::numeric_limits<uint16_t>::max()) + 2 };
for (size_t max_num_bin : max_num_bins) {
auto dmat = RandomDataGenerator(100, 1, 0.5).GenerateDMatrix();
GHistIndexMatrix gmat(dmat.get(), max_num_bin, false);
GHistIndexMatrix gmat(dmat.get(), max_num_bin, false, common::OmpGetNumThreads(0));
ColumnMatrix column_matrix;
column_matrix.Init(gmat, 0.2);
switch (column_matrix.GetTypeSize()) {
@@ -130,7 +133,7 @@ void TestGHistIndexMatrixCreation(size_t nthreads) {
/* This should create multiple sparse pages */
std::unique_ptr<DMatrix> dmat{ CreateSparsePageDMatrix(kEntries) };
omp_set_num_threads(nthreads);
GHistIndexMatrix gmat(dmat.get(), 256, false);
GHistIndexMatrix gmat(dmat.get(), 256, false, common::OmpGetNumThreads(0));
}
TEST(HistIndexCreationWithExternalMemory, Test) {

42
tests/cpp/common/test_hist_util.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2019-2021 by XGBoost Contributors
* Copyright 2019-2022 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <vector>
@@ -188,7 +188,7 @@ TEST(HistUtil, DenseCutsCategorical) {
std::vector<float> x_sorted(x);
std::sort(x_sorted.begin(), x_sorted.end());
auto dmat = GetDMatrixFromData(x, n, 1);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
auto cuts_from_sketch = cuts.Values();
EXPECT_LT(cuts.MinValues()[0], x_sorted.front());
EXPECT_GT(cuts_from_sketch.front(), x_sorted.front());
@@ -207,7 +207,7 @@ TEST(HistUtil, DenseCutsAccuracyTest) {
auto x = GenerateRandom(num_rows, num_columns);
auto dmat = GetDMatrixFromData(x, num_rows, num_columns);
for (auto num_bins : bin_sizes) {
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
ValidateCuts(cuts, dmat.get(), num_bins);
}
}
@@ -224,11 +224,13 @@ TEST(HistUtil, DenseCutsAccuracyTestWeights) {
dmat->Info().weights_.HostVector() = w;
for (auto num_bins : bin_sizes) {
{
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, true);
HistogramCuts cuts =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), true);
ValidateCuts(cuts, dmat.get(), num_bins);
}
{
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, false);
HistogramCuts cuts =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), false);
ValidateCuts(cuts, dmat.get(), num_bins);
}
}
@@ -249,13 +251,15 @@ void TestQuantileWithHessian(bool use_sorted) {
dmat->Info().weights_.HostVector() = w;
for (auto num_bins : bin_sizes) {
HistogramCuts cuts_hess = SketchOnDMatrix(dmat.get(), num_bins, use_sorted, hessian);
HistogramCuts cuts_hess =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), use_sorted, hessian);
for (size_t i = 0; i < w.size(); ++i) {
dmat->Info().weights_.HostVector()[i] = w[i] * hessian[i];
}
ValidateCuts(cuts_hess, dmat.get(), num_bins);
HistogramCuts cuts_wh = SketchOnDMatrix(dmat.get(), num_bins, use_sorted);
HistogramCuts cuts_wh =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), use_sorted);
ValidateCuts(cuts_wh, dmat.get(), num_bins);
ASSERT_EQ(cuts_hess.Values().size(), cuts_wh.Values().size());
@@ -283,7 +287,7 @@ TEST(HistUtil, DenseCutsExternalMemory) {
auto dmat =
GetExternalMemoryDMatrixFromData(x, num_rows, num_columns, 50, tmpdir);
for (auto num_bins : bin_sizes) {
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
ValidateCuts(cuts, dmat.get(), num_bins);
}
}
@@ -303,7 +307,7 @@ TEST(HistUtil, IndexBinBound) {
for (auto max_bin : bin_sizes) {
auto p_fmat = RandomDataGenerator(kRows, kCols, 0).GenerateDMatrix();
GHistIndexMatrix hmat(p_fmat.get(), max_bin, false);
GHistIndexMatrix hmat(p_fmat.get(), max_bin, false, common::OmpGetNumThreads(0));
EXPECT_EQ(hmat.index.Size(), kRows*kCols);
EXPECT_EQ(expected_bin_type_sizes[bin_id++], hmat.index.GetBinTypeSize());
}
@@ -326,7 +330,7 @@ TEST(HistUtil, IndexBinData) {
for (auto max_bin : kBinSizes) {
auto p_fmat = RandomDataGenerator(kRows, kCols, 0).GenerateDMatrix();
GHistIndexMatrix hmat(p_fmat.get(), max_bin, false);
GHistIndexMatrix hmat(p_fmat.get(), max_bin, false, common::OmpGetNumThreads(0));
uint32_t* offsets = hmat.index.Offset();
EXPECT_EQ(hmat.index.Size(), kRows*kCols);
switch (max_bin) {
@@ -351,7 +355,7 @@ void TestSketchFromWeights(bool with_group) {
size_t constexpr kGroups = 10;
auto m =
RandomDataGenerator{kRows, kCols, 0}.Device(0).GenerateDMatrix();
common::HistogramCuts cuts = SketchOnDMatrix(m.get(), kBins);
common::HistogramCuts cuts = SketchOnDMatrix(m.get(), kBins, common::OmpGetNumThreads(0));
MetaInfo info;
auto& h_weights = info.weights_.HostVector();
@@ -385,7 +389,7 @@ void TestSketchFromWeights(bool with_group) {
ValidateCuts(cuts, m.get(), kBins);
if (with_group) {
HistogramCuts non_weighted = SketchOnDMatrix(m.get(), kBins);
HistogramCuts non_weighted = SketchOnDMatrix(m.get(), kBins, common::OmpGetNumThreads(0));
for (size_t i = 0; i < cuts.Values().size(); ++i) {
EXPECT_EQ(cuts.Values()[i], non_weighted.Values()[i]);
}
@@ -404,14 +408,12 @@ TEST(HistUtil, SketchFromWeights) {
}
TEST(HistUtil, SketchCategoricalFeatures) {
TestCategoricalSketch(1000, 256, 32, false,
[](DMatrix *p_fmat, int32_t num_bins) {
return SketchOnDMatrix(p_fmat, num_bins);
});
TestCategoricalSketch(1000, 256, 32, true,
[](DMatrix *p_fmat, int32_t num_bins) {
return SketchOnDMatrix(p_fmat, num_bins);
});
TestCategoricalSketch(1000, 256, 32, false, [](DMatrix* p_fmat, int32_t num_bins) {
return SketchOnDMatrix(p_fmat, num_bins, common::OmpGetNumThreads(0));
});
TestCategoricalSketch(1000, 256, 32, true, [](DMatrix* p_fmat, int32_t num_bins) {
return SketchOnDMatrix(p_fmat, num_bins, common::OmpGetNumThreads(0));
});
}
} // namespace common
} // namespace xgboost

6
tests/cpp/common/test_hist_util.cu
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2019-2021 by XGBoost Contributors
* Copyright 2019-2022 by XGBoost Contributors
*/
#include <dmlc/filesystem.h>
#include <gtest/gtest.h>
@@ -28,7 +28,7 @@ namespace common {
template <typename AdapterT>
HistogramCuts GetHostCuts(AdapterT *adapter, int num_bins, float missing) {
data::SimpleDMatrix dmat(adapter, missing, 1);
HistogramCuts cuts = SketchOnDMatrix(&dmat, num_bins);
HistogramCuts cuts = SketchOnDMatrix(&dmat, num_bins, common::OmpGetNumThreads(0));
return cuts;
}
@@ -40,7 +40,7 @@ TEST(HistUtil, DeviceSketch) {
auto dmat = GetDMatrixFromData(x, num_rows, num_columns);
auto device_cuts = DeviceSketch(0, dmat.get(), num_bins);
HistogramCuts host_cuts = SketchOnDMatrix(dmat.get(), num_bins);
HistogramCuts host_cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
EXPECT_EQ(device_cuts.Values(), host_cuts.Values());
EXPECT_EQ(device_cuts.Ptrs(), host_cuts.Ptrs());

5
tests/cpp/common/test_quantile.cc
View File

@@ -1,3 +1,6 @@
/*!
* Copyright 2020-2022 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include "test_quantile.h"
#include "../../../src/common/quantile.h"
@@ -201,7 +204,7 @@ TEST(Quantile, SameOnAllWorkers) {
.MaxCategory(17)
.Seed(rank + seed)
.GenerateDMatrix();
auto cuts = SketchOnDMatrix(m.get(), n_bins);
auto cuts = SketchOnDMatrix(m.get(), n_bins, common::OmpGetNumThreads(0));
std::vector<float> cut_values(cuts.Values().size() * world, 0);
std::vector<
typename std::remove_reference_t<decltype(cuts.Ptrs())>::value_type>

5
tests/cpp/data/test_data.cc
View File

@@ -1,3 +1,6 @@
/*!
* Copyright 2019-2022 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <dmlc/filesystem.h>
#include <fstream>
@@ -66,7 +69,7 @@ TEST(SparsePage, PushCSCAfterTranspose) {
SparsePage page; // Consolidated sparse page
for (const auto &batch : dmat->GetBatches<xgboost::SparsePage>()) {
// Transpose each batch and push
SparsePage tmp = batch.GetTranspose(ncols);
SparsePage tmp = batch.GetTranspose(ncols, common::OmpGetNumThreads(0));
page.PushCSC(tmp);
}

4
tests/cpp/data/test_gradient_index.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2021 XGBoost contributors
* Copyright 2021-2022 XGBoost contributors
*/
#include <gtest/gtest.h>
#include <xgboost/data.h>
@@ -36,7 +36,7 @@ TEST(GradientIndex, FromCategoricalBasic) {
BatchParam p(0, max_bins);
GHistIndexMatrix gidx;
gidx.Init(m.get(), max_bins, false, {});
gidx.Init(m.get(), max_bins, false, common::OmpGetNumThreads(0), {});
auto x_copy = x;
std::sort(x_copy.begin(), x_copy.end());

5
tests/cpp/tree/hist/test_evaluate_splits.cc
View File

@@ -1,3 +1,6 @@
/*!
* Copyright 2021-2022 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <xgboost/base.h>
#include "../../../../src/tree/hist/evaluate_splits.h"
@@ -29,7 +32,7 @@ template <typename GradientSumT> void TestEvaluateSplits() {
size_t constexpr kMaxBins = 4;
// dense, no missing values
GHistIndexMatrix gmat(dmat.get(), kMaxBins, false);
GHistIndexMatrix gmat(dmat.get(), kMaxBins, false, common::OmpGetNumThreads(0));
common::RowSetCollection row_set_collection;
std::vector<size_t> &row_indices = *row_set_collection.Data();
row_indices.resize(kRows);

8
tests/cpp/tree/test_quantile_hist.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2018-2021 by Contributors
* Copyright 2018-2022 by XGBoost Contributors
*/
#include <xgboost/host_device_vector.h>
#include <xgboost/tree_updater.h>
@@ -162,7 +162,7 @@ class QuantileHistMock : public QuantileHistMaker {
// kNRows samples with kNCols features
auto dmat = RandomDataGenerator(kNRows, kNCols, sparsity).Seed(3).GenerateDMatrix();
GHistIndexMatrix gmat(dmat.get(), kMaxBins, false);
GHistIndexMatrix gmat(dmat.get(), kMaxBins, false, common::OmpGetNumThreads(0));
ColumnMatrix cm;
// treat everything as dense, as this is what we intend to test here
@@ -253,7 +253,7 @@ class QuantileHistMock : public QuantileHistMaker {
void TestInitData() {
size_t constexpr kMaxBins = 4;
GHistIndexMatrix gmat(dmat_.get(), kMaxBins, false);
GHistIndexMatrix gmat(dmat_.get(), kMaxBins, false, common::OmpGetNumThreads(0));
RegTree tree = RegTree();
tree.param.UpdateAllowUnknown(cfg_);
@@ -270,7 +270,7 @@ class QuantileHistMock : public QuantileHistMaker {
void TestInitDataSampling() {
size_t constexpr kMaxBins = 4;
GHistIndexMatrix gmat(dmat_.get(), kMaxBins, false);
GHistIndexMatrix gmat(dmat_.get(), kMaxBins, false, common::OmpGetNumThreads(0));
RegTree tree = RegTree();
tree.param.UpdateAllowUnknown(cfg_);